Subject statistics

First, we will determine whether IQ correlates with age in our sample. As specified in our pre-registration, here we run run a linear regression testing the effects of age on IQ.

Analysis of Variance Table
  Df Sum Sq Mean Sq F value Pr(>F) pEta
Age 1 1375 1375 8.642 0.004194 0.08943
Residuals 88 14005 159.1 NA NA NA

This analysis indicates that there is a significant relation between age and IQ. As such, we will run all subsequent models both with and without IQ as an interacting fixed effect to better tease apart effects of age from effects of IQ.

Frequency task analysis

In this section, we will run mixed effects models to determine the effects of appearance count and age on participants’ accuracy and reaction time during the frequency task. For both models, we will start with the maximal random effects structure. If the models do not converge, we will remove random slopes and intercepts as specified in the pre-registration. As we preregistered, if the maximal model does not converge, we will follow the procedure laid out in Singmann & Kellen (2017) to reach convergence. They recommend first removing correlations between random slopes and random intercepts. If models stil fail to converge, they recommend removing random-effects parameters ’starting with the highest-order random effects parameter with the lowest estimated variance.

Frequency task accuracy

First, we will ensure that all subjects responded to >10% of repeated trials, indicating that they understood the task instructions.

Frequency accuracy mixed-effects model: New items without IQ

## Fitting 2 (g)lmer() models:
## [..]
## Mixed Model Anova Table (Type 3 tests, LRT-method)
## 
## Model: acc ~ ageScaled + (1 | ID) + (ageScaled | animal)
## Data: freqDataNewItems
## Df full model: 6
##      Effect df     Chisq p.value
## 1 ageScaled  1 19.92 ***  <.0001
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Frequency accuracy mixed-effects model: New items with IQ

## Fitting 4 (g)lmer() models:
## [....]
## Mixed Model Anova Table (Type 3 tests, LRT-method)
## 
## Model: acc ~ ageScaled * IQScaled + (1 | ID) + (ageScaled * IQScaled | 
## Model:     animal)
## Data: freqDataNewItems
## Df full model: 15
##               Effect df     Chisq p.value
## 1          ageScaled  1 18.24 ***  <.0001
## 2           IQScaled  1      0.61     .43
## 3 ageScaled:IQScaled  1    5.14 *     .02
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Frequency accuracy: Repeated items without IQ

## Fitting 4 (g)lmer() models:
## [....]
## Mixed Model Anova Table (Type 3 tests, LRT-method)
## 
## Model: acc ~ ageScaled * appearanceCountScaled + (appearanceCountScaled | 
## Model:     ID) + (ageScaled * appearanceCountScaled | animal)
## Data: freqDataRepeatedItems
## Df full model: 17
##                            Effect df     Chisq p.value
## 1                       ageScaled  1   6.71 **    .010
## 2           appearanceCountScaled  1 54.93 ***  <.0001
## 3 ageScaled:appearanceCountScaled  1    3.54 +     .06
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Frequency accuracy: Repeated items with IQ

## Fitting 8 (g)lmer() models:
## [........]
## Mixed Model Anova Table (Type 3 tests, LRT-method)
## 
## Model: acc ~ ageScaled * IQScaled * appearanceCountScaled + (appearanceCountScaled | 
## Model:     ID) + (ageScaled * IQScaled * appearanceCountScaled | animal)
## Data: freqDataRepeatedItems
## Df full model: 47
##                                     Effect df     Chisq p.value
## 1                                ageScaled  1   7.30 **    .007
## 2                                 IQScaled  1  10.24 **    .001
## 3                    appearanceCountScaled  1 48.99 ***  <.0001
## 4                       ageScaled:IQScaled  1      0.09     .77
## 5          ageScaled:appearanceCountScaled  1      1.14     .29
## 6           IQScaled:appearanceCountScaled  1      0.91     .34
## 7 ageScaled:IQScaled:appearanceCountScaled  1      1.43     .23
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Frequency accuracy: Plot

Frequency task reaction times: Repeated items without IQ

## Fitting one lmer() model. [DONE]
## Calculating p-values. [DONE]
## Mixed Model Anova Table (Type 3 tests, KR-method)
## 
## Model: freqRT ~ appearanceCountScaled * ageScaled + (appearanceCountScaled | 
## Model:     ID) + (appearanceCountScaled * ageScaled | animal)
## Data: freqRTdata
##                            Effect       df          F p.value
## 1           appearanceCountScaled 1, 70.06 103.09 ***  <.0001
## 2                       ageScaled 1, 91.59   10.63 **    .002
## 3 appearanceCountScaled:ageScaled 1, 70.90     6.49 *     .01
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Frequency task reaction times: Repeated items with IQ

## Fitting one lmer() model. [DONE]
## Calculating p-values. [DONE]
## Mixed Model Anova Table (Type 3 tests, KR-method)
## 
## Model: freqRT ~ appearanceCountScaled * ageScaled * IQScaled + (appearanceCountScaled | 
## Model:     ID) + (appearanceCountScaled * ageScaled * IQScaled | animal)
## Data: freqRTdata
##                                     Effect       df         F p.value
## 1                    appearanceCountScaled 1, 69.72 90.58 ***  <.0001
## 2                                ageScaled 1, 89.98 14.25 ***   .0003
## 3                                 IQScaled 1, 89.71    4.44 *     .04
## 4          appearanceCountScaled:ageScaled 1, 70.32   7.62 **    .007
## 5           appearanceCountScaled:IQScaled 1, 70.10      0.77     .38
## 6                       ageScaled:IQScaled 1, 93.57      0.28     .60
## 7 appearanceCountScaled:ageScaled:IQScaled 1, 69.85      0.65     .42
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Frequency task reaction times: Plot

Memory test analysis

In this section, we will run four mixed effects models to examine the effects of frequency condition on memory accuracy (both with and without IQ included). We will deviate from our pre-registration by only analyzing the data from the first appearance of each item during the memory test (the first 30 trials). We will follow the same procedure as with our frequency data to find the maximal random effects structure that converges.

Memory accuracy without IQ

## Fitting 4 (g)lmer() models:
## [....]
## Mixed Model Anova Table (Type 3 tests, LRT-method)
## 
## Model: memAcc ~ freqCondScaled * ageScaled + (freqCondScaled || ID) + 
## Model:     (freqCondScaled * ageScaled || animal)
## Data: memDataCombined
## Df full model: 10
##                     Effect df   Chisq p.value
## 1           freqCondScaled  1    0.51     .48
## 2                ageScaled  1 9.15 **    .002
## 3 freqCondScaled:ageScaled  1    0.02     .89
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Memory accuracy with IQ

## Fitting 8 (g)lmer() models:
## [........]
## Mixed Model Anova Table (Type 3 tests, LRT-method)
## 
## Model: memAcc ~ freqCondScaled * ageScaled * IQScaled + (freqCondScaled || 
## Model:     ID) + (freqCondScaled * ageScaled * IQScaled || animal)
## Data: memDataCombined
## Df full model: 18
##                              Effect df     Chisq p.value
## 1                    freqCondScaled  1      0.01     .93
## 2                         ageScaled  1 14.69 ***   .0001
## 3                          IQScaled  1 11.18 ***   .0008
## 4          freqCondScaled:ageScaled  1      0.17     .68
## 5           freqCondScaled:IQScaled  1      0.59     .44
## 6                ageScaled:IQScaled  1      0.27     .60
## 7 freqCondScaled:ageScaled:IQScaled  1    3.50 +     .06
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '+' 0.1 ' ' 1

Relationship between structural learning and memory encoding

As specified, we will now run a regression examining the effects of structural learning index on memory benefit score, to determine if participants who showed the most evidence of learning the animal frequencies also demonstrated the greatest effect of frequency condition on encoding.

Structural learning model without IQ

  Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.09911 0.1035 -0.9578 0.3409
ageScaled 0.03167 0.1044 0.3034 0.7623
learningIndexScaled 0.007472 0.1064 0.07022 0.9442
ageScaled:learningIndexScaled 0.01561 0.09922 0.1573 0.8753
Fitting linear model: memBenefitIndex ~ ageScaled * learningIndexScaled
Observations Residual Std. Error \(R^2\) Adjusted \(R^2\)
89 0.9726 0.001356 -0.03389

Structural learning model with IQ

Table continues below
  Estimate Std. Error t value
(Intercept) -0.03462 0.1069 -0.3238
ageScaled 0.08747 0.1116 0.7838
IQScaled 0.0485 0.1104 0.4392
learningIndexScaled 0.012 0.1099 0.1092
ageScaled:IQScaled 0.248 0.1073 2.311
ageScaled:learningIndexScaled -0.007685 0.1027 -0.07484
IQScaled:learningIndexScaled 0.06565 0.1144 0.5736
ageScaled:IQScaled:learningIndexScaled 0.05675 0.1117 0.5081
  Pr(>|t|)
(Intercept) 0.7469
ageScaled 0.4355
IQScaled 0.6617
learningIndexScaled 0.9133
ageScaled:IQScaled 0.02336
ageScaled:learningIndexScaled 0.9405
IQScaled:learningIndexScaled 0.5678
ageScaled:IQScaled:learningIndexScaled 0.6127
Fitting linear model: memBenefitIndex ~ ageScaled * IQScaled * learningIndexScaled
Observations Residual Std. Error \(R^2\) Adjusted \(R^2\)
89 0.9611 0.07074 -0.009569

Plot: Correlation between frequency learning and memory

Plot: Interaction between age and IQ on memory benefit score

Relationship between strategy use and memory encoding

Finally, we will run a regression examining the effects of strategy use on memory benefit score. We will model the effect of subject strategy recognition (a 2-level categorical variable based on participant’s response (yes/no) to our question of whether they recognized that focusing on the more-frequent animals would be a good strategy) and age on their memory benefit score.

Strategy use regression without IQ

  Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.6771 0.2296 -2.949 0.004117
ageScaled 0.03278 0.2102 0.156 0.8764
GeneratedStrategy1 0.7153 0.2545 2.811 0.006128
ageScaled:GeneratedStrategy1 -0.04419 0.2385 -0.1853 0.8535
Fitting linear model: memBenefitIndex ~ ageScaled * GeneratedStrategy
Observations Residual Std. Error \(R^2\) Adjusted \(R^2\)
89 0.9288 0.08929 0.05715

Strategy use regression with IQ

Table continues below
  Estimate Std. Error t value
(Intercept) -0.4431 0.2677 -1.655
ageScaled 0.2314 0.2609 0.8868
IQScaled -0.002351 0.2269 -0.01036
GeneratedStrategy1 0.5094 0.2902 1.755
ageScaled:IQScaled 0.4505 0.2165 2.081
ageScaled:GeneratedStrategy1 -0.2019 0.286 -0.7058
IQScaled:GeneratedStrategy1 0.07963 0.256 0.311
ageScaled:IQScaled:GeneratedStrategy1 -0.3265 0.2473 -1.32
  Pr(>|t|)
(Intercept) 0.1017
ageScaled 0.3778
IQScaled 0.9918
GeneratedStrategy1 0.083
ageScaled:IQScaled 0.04058
ageScaled:GeneratedStrategy1 0.4823
IQScaled:GeneratedStrategy1 0.7566
ageScaled:IQScaled:GeneratedStrategy1 0.1904
Fitting linear model: memBenefitIndex ~ ageScaled * IQScaled * GeneratedStrategy
Observations Residual Std. Error \(R^2\) Adjusted \(R^2\)
89 0.9123 0.1627 0.09031

Strategy use by age group

ageGroup meanStratUse
Child 0.7667
Adolescent 0.8333
Adult 0.8333